The chief executive of AI company Anthropic has called for governments to have the authority to stop dangerous artificial intelligence systems, arguing that powerful models should be subject to risk-based oversight.
Dario Amodei said in a blog post published on Wednesday that advanced AI systems should be evaluated for risks including cybersecurity threats, biological weapons misuse and the potential loss of control over the systems, including through automated self-improvement that could amplify such risks.
Such assessments could be carried out by a government body similar to an aviation regulator, or by government-appointed private auditors, he said.
Anthropic has long advocated tighter regulation of the AI industry and has faced criticism from parts of the sector as well as from the White House for its stance.
The company, a major competitor to ChatGPT developer OpenAI, recently drew attention with a model capable of identifying long-standing vulnerabilities in widely used software, heightening concerns that AI could be used to enable large-scale cyberattacks.
